personalagentbench/personal-agent-bench

Personal Agent Bench

Task completion rates for personal AI agents, published with 95% intervals and sample sizes, plus the runtime teardown behind each product. This is the data from personalagentbench.com; the pages there explain every figure.

There is no combined score and no ranking. A rate without its interval and the number of tasks it rests on is not a result, so every row carries both.

Files

• results.csv: one row per product per batch. Columns: batch, as_of, product, slug, status, rate_pct, ci95_low_pct, ci95_high_pct, scored_tasks, runs, same_tasks_for_all, note.

• results.json: the same figures with each batch's description and caveat. Read the caveat before comparing batches; the September and October batches used different tasks, scoring rules and exam versions.

• scorecard.json: what we found when we took each runtime apart, on these questions:

• execution: Where it runs. Does the model run near your machine or far from it, and what can it touch?

• tools: How tools are exposed. When does a tool definition enter the context, and who decides which ones?

• memory: Where memory lives. Files you can open, or a server-side store you can only reach through the product?

• control: What you get to approve. What does it ask before acting, and can you take back what it knows?

• channel: How you reach it. What is the default surface, and what happens on the ones that are not?

• unknown: What we could not establish. Stated plainly, because a bench that only reports wins is not a bench.

Batches

Batches are listed separately and never added together. Each batch's note in results.json says what it can and cannot be compared with.

Batch	Label	As of	Same tasks for every product
october	October batch (11 October)	2026-10-11	yes, 20
october-other-run-1014b-main	Other October batch (14 October, second run)	2026-10-14	yes, 7
october-other-run-1014-main	Other October batch (14 October, first run)	2026-10-14	yes, 7
october-other-run-1012-main	Other October batch (12 October, large batch)	2026-10-12	yes, 72
october-other-takeover-1011-pilot-paid-rerun	Other October batch (11 October, rerun)	2026-10-11	yes, 20
october-other-pilot-1011-batch-v2	Other October batch (11 October, pilot)	2026-10-11	yes, 20
first-batch	First batch	2026-09-22	no
same-tasks	The 09-21 to 09-22 records, filtered	2026-09-22	yes, 114

Method and provenance

Every figure is a share of tasks completed, with a 95% interval and the number of scored tasks it rests on. There is no combined score and no ranking, because a rate without its interval and its sample size is not a result. These September rates describe adapted runtimes using the exam model, on differing scored subsets; they are not consumer-product end-to-end performance.

On the September batches: these figures come from the team running the exam. The bench repository deliberately stopped tracking generated batch records in September, on the grounds that it holds the inputs a run needs rather than the outputs a run produces, so there is no published record behind these numbers for us or anyone else to check them against. We are printing them anyway, with that said plainly, because the alternative is printing nothing and pretending we have no numbers. If the underlying batches are ever published we will reconcile against them and say what changed. The October batches differ: the note on each says whether its figures were checked against the records behind them.

How the exam works: https://personalagentbench.com/how-the-exam-works/. How scoring works: https://personalagentbench.com/how-scoring-works/. Changes to the exam: https://personalagentbench.com/how-the-exam-works/changes/.

The exam tasks themselves are not in this dataset, so that they stay usable as a test.

Who runs it

Published by Personal Agent Bench, run by the Zentor team. Zentor builds agent software of its own, which is a conflict of interest, so its products are not on the scorecard and will not be added.

Corrections

Every corrected claim is logged at https://personalagentbench.com/corrections/ (machine-readable: https://personalagentbench.com/corrections.json). This copy was exported on 2026-10-22; the latest batch in it is from 2026-10-14. The website is the current version.

Licence and citation

CC BY 4.0. Please credit Personal Agent Bench and link to https://personalagentbench.com/results/.

@misc{personalagentbench,
  title  = {Personal Agent Bench},
  author = {{Personal Agent Bench}},
  year   = {2026},
  url    = {https://personalagentbench.com/results/}
}